A formula to guide decision making compares the cost of allowing a failure to occur versus the cost to proactively repair the component while considering the probability of predicting the failure:

If the cost of failure is the same as the proactive repair cost, even with a perfect prediction model, which we know from Chapter 10, Data Science for IoT Analytics is highly unlikely, then there will be no savings. Make sure to include intangible costs into the cost of failure. Some examples of intangible costs include legal expenses, loss of brand equity, and even the customer's expenses.

Predictive repair does makes sense when there is a large spread between the cost of failure and the cost of proactive replacement, combined with a well-performing prediction model. For example, if the cost of a failure is a locomotive engine replacement at $1 million USD and the cost of a proactive repair is $200 USD, then the accuracy of the model does not even have to be all that great before a proactive replacement program makes financial sense.

On the other hand, if the failure is a $400 USD automotive turbocharger replacement, and the proactive repair cost is $350 USD for a turbocharger actuator subcomponent replacement, the predictive model would need to be highly accurate for that to make financial sense.